A method and system for detecting carbon emissions from enterprise loads

Abstract

The present invention provides a method and system for detecting carbon emissions from enterprise loads, and belongs to the technical field of enterprise load carbon emissions detection. The technical problem to be solved is: to provide an improvement on a method and system structure for detecting carbon emissions from enterprise loads. The technical solution adopted to solve the technical problem is: to analyze the direct carbon emissions of an enterprise by establishing an enterprise load equipment operation status model, to model photovoltaics, to adopt an equivalent output estimation model to reduce its influence on the accounting results, to calculate the indirect carbon emissions of the enterprise by constructing an improved carbon emission flow model, to establish a comprehensive carbon emissions calculation model, to calculate the total carbon emissions including direct and indirect carbon emissions, and finally to achieve the purpose of calculating the carbon emissions of industrial production. The present invention is applied to the detection and calculation of carbon emissions from enterprise loads.

Description

Technical Field

[0001] The present invention provides a method and system for detecting carbon emissions from enterprise loads, belonging to the technical field of detecting carbon emissions from enterprise loads. Background Art

[0002] In recent years, with the continuous development of society, industrial energy consumption, greenhouse gas emissions and pollution have also increased, resulting in frequent extreme weather events and natural disasters causing casualties and economic losses. Therefore, implementing a low-carbon economy and gradually moving towards carbon neutrality is a development trend and is also very urgent. At present, the main means to achieve carbon neutrality is to reduce and control the carbon emissions of enterprises. Therefore, it is of great significance to accurately calculate the carbon emissions of enterprises and make corresponding production plans based on accurate carbon emissions.

[0003] At present, the calculation of corporate carbon emissions is mainly based on corporate activity data: some studies are based on input-output analysis, using financial data and input-output data to evaluate the impact of carbon emissions; some studies use life cycle assessment methods to evaluate carbon dioxide emissions in the production process of different industries; some studies use financial accounting methods to analyze carbon emissions by considering financial, economic and other factors; however, the above methods all have certain defects and shortcomings. First, the carbon emissions they calculate are carbon emissions after a period of production, which has a certain hysteresis; secondly, companies usually need to keep financial data confidential during actual operations, and it is often difficult to obtain valid data when using the above methods to calculate carbon emissions; therefore, there is an urgent need to propose a method that relies less on sensitive data and can accurately detect and calculate carbon emissions.

[0004] At the same time, with the widespread application of photovoltaic systems, the measurement of actual corporate loads has been further challenged. Photovoltaic systems are usually connected behind smart meters, resulting in the actual measurement value of smart meters being the net load of corporate load demand minus photovoltaic output, causing the actual output of photovoltaic systems to be ignored and the actual corporate load to be hidden. On the other hand, corporate electricity demand and photovoltaic output are volatile. Usually, the actual load demand is greater than photovoltaic power generation. However, when photovoltaic power generation is greater than load demand, the company will transmit electricity to the power system as a generator, that is, the company has two identities as a consumer and a "generator", which will ultimately lead to further increases in the difficulty of carbon emission detection and calculation for the company. Summary of the invention

[0005] In order to overcome the deficiencies in the prior art, the present invention aims to solve the following technical problems: to provide an enterprise load carbon emission detection method and an improvement in the detection system structure.

[0006] In order to solve the above technical problems, the technical solution adopted by the present invention is: a method for detecting carbon emissions of enterprise load, comprising the following detection steps:

[0007] S1: Initialization, inputting parameters related to enterprise load carbon emissions;

[0008] S2: Introduce and improve the Informer algorithm, establish a DOS estimation model for the operating status of enterprise load equipment, and analyze the direct carbon emissions of enterprises;

[0009] S3: Use the general physical photovoltaic model to approximate the photovoltaic output effect and establish a photovoltaic system model;

[0010] S4: Construct an improved carbon emission flow CEF model to calculate the indirect carbon emissions of enterprises;

[0011] S5: Establish a comprehensive carbon emissions calculation model to calculate the total carbon emissions including direct carbon emissions and indirect carbon emissions.

[0012] The enterprise load equipment operation status DOS estimation model established in step S2 includes data preparation, model training and DOS estimation. The data includes the net load of the enterprise and the DOS corresponding to each device. The specific modeling method is:

[0013] S21: In the data preparation stage, min-max normalization is used to convert the net load P NL,t The value is scaled to between 0 and 1, where the net load P NL,t is the actual load P L,t With photovoltaic output P PV,t The difference is expressed as:

[0014] P NL,t =P L,t -P PV,t ;

[0015] For a device, there are training load data X, test load data X' and predicted load data X'', which are K1, K2 and K3 dimensional vectors respectively; corresponding to the data X, X' and X'', the training, test and predicted DOS data are Y, Y' and Y'' respectively, which are used as category labels and converted into binary matrices of dimensions m×K1, m×K2 and m×K3 respectively, where m represents the number of operating states of the device;

[0016] S22: Use the improved Informer algorithm to estimate DOS when training the model, and modify the initial Informer model into a multi-class prediction model;

[0017] S23: Substitute the predicted load data X'' into the trained model to obtain the predicted DOS data Y''.

[0018] The Informer algorithm introduced and improved in step S2 includes ProbSparse self-attention mechanism, self-attention extraction and generative decoder;

[0019] The improvements to the Informer model include: changing the target data to the category label of DOS, selecting the CrossEntropyLoss function as the loss function, setting the ReLU function as the activation function, and adding a dense layer at the end of the model.

[0020] The specific method of using the general physical photovoltaic model to approximate the photovoltaic output effect in step S3 is:

[0021] The photovoltaic panels are set as a virtual equivalent photovoltaic system with a fixed tilt photovoltaic array;

[0022] Inputs include weather data, solar irradiance data, and PV system technical parameters;

[0023] The PV output is approximated using a general physical PV model, and the PV system technical parameters that need to be estimated include: the capacity C of the PV power plant, the tilt angle β, and the orientation γ of the PV panels, which are decision variables;

[0024] Let D be the total number of devices, η d is the DOS estimation accuracy of device d, K is the total number of data samples; binary variable σ d,t is the accuracy of the estimated DOS at time t. If the estimated DOS value of device d at time t is correct, then σ d,t is equal to 1, otherwise σ d,t Equal to 0; o d,t 、o' d,t are the actual and estimated DOS of device d at time t respectively; the PV output is indirectly estimated through the DOS estimation accuracy, and the objective function is to minimize the sum of DOS estimation errors, expressed as:

[0025] ;

[0026] Let I PV,t is the solar radiation received by the photovoltaic panel at time t, T A,t is the outdoor ambient temperature at time t, N oct is the rated working unit temperature, then the temperature of the photovoltaic module at time t is T PV,t Approximate expression is:

[0027] ;

[0028] Assuming μ is the temperature coefficient, the photovoltaic output P PV,t Approximate expression is:

[0029] .

[0030] The specific method for establishing the photovoltaic system model in step S3 is:

[0031] Let I 0,t is the direct normal irradiance at time t; τ b,t is the atmospheric transparency coefficient at time t; I b,t ,I h,t They represent the direct reflection horizontal irradiance and diffuse reflection horizontal irradiance at time t respectively; ρ t is the surface albedo at time t, α is the height of the sun, A is the direction of the sun, then the solar radiation I received by the photovoltaic panel at time t PV,t It is the sum of direct reflection horizontal irradiance, diffuse reflection horizontal irradiance and direct reflection horizontal irradiance at time t, and the expression is:

[0032] .

[0033] The specific method for constructing the improved carbon emission flow CEF model in step S4 is:

[0034] Calculate the real-time node CEI, assuming P NL,t is the node net load column vector at time t, e N,t is the column vector of the CEI of the node electricity at time t, then the carbon emission vector of the node electricity at time t is The expression is:

[0035] ;

[0036] in, and P NL,t They are all N-dimensional column vectors, and Δt is the time interval.

[0037] The specific method for calculating the indirect carbon emissions of the enterprise in step S4 is:

[0038] Assume R C,t is the column vector of carbon emission flow released at time t, e N,t and R C,t are all N-dimensional column vectors; P NB,t Calculate the discriminant matrix for CEF at time t, P B,out,t is the branch output power distribution matrix at time t, P NB,t and P B,out,t are all N-dimensional square matrices; t is the unit (N+G+P)-dimensional row vector; P G,t is the power injection distribution matrix at time t, which is an N×(G+P)-dimensional column vector; E G,t is the column vector of the power CEI at time t, which is a (G+P)-dimensional column vector; P gen,tis the power injection distribution matrix of the generator at time t, which is an N×G dimensional matrix; P pro,t is the power injection distribution matrix of the prosumer at time t, which is an N×P dimensional matrix; E gen,t is the column vector of producer CEI at time t, which is a G-dimensional column vector; E pro,t is the column vector of the CEI of the prosumer at time t, which is a P-dimensional column vector; "diag{}" is a diagonal matrix operator; considering CEF as a virtual flow together with the power flow, for the n-node power system at time t, the expression of the real-time node CEI is:

[0039] .

[0040] The specific method for establishing the comprehensive carbon emission calculation model and calculating the total carbon emission in step S5 is:

[0041] Assume E k is the total carbon emissions of node k; E k,scope1 is the direct carbon emission of node k; E k,scope2 is the indirect carbon emission of node k; K is the total number of data samples; D is the total number of devices; E device,d,t is the direct carbon emission of device d at time t; is the carbon emission of electricity at node k at time t; e d,t is the CEI of device d at time t; o' d,t is the estimated DOS of device d at time t; constant value e d1 To e dm is the CEI of equipment d under different working conditions; then the expression of the total carbon emissions of the enterprise at node k is:

[0042] .

[0043] A detection system used in a method for detecting carbon emissions from enterprise loads includes the following detection devices:

[0044] The main server is responsible for the management and scheduling of the overall carbon emissions calculation task and has data processing and storage functions;

[0045] Computing nodes: Multiple computing nodes are connected to the main server through the network to form a distributed computing cluster. Each computing node is responsible for executing specific carbon emission calculation tasks.

[0046] The control terminal is used for user interaction and has the functions of inputting calculation parameters and viewing calculation results. The control terminal is connected to the main server through a local area network or the Internet;

[0047] The data acquisition terminal is responsible for collecting real-time load data and weather data of the enterprise power grid and sending the data to the main server through the data transmission channel.

[0048] The main server includes:

[0049] The data preprocessing module is used to receive the enterprise net load, DOS data and meteorological data from the data acquisition terminal, and perform data cleaning, format conversion, abnormal value detection and processing;

[0050] A construction module is used to obtain the carbon emission data of enterprises and establish a comprehensive carbon emission calculation model that considers direct and indirect carbon emissions. Specifically, the DOS estimation model is constructed using the improved Informer algorithm, the photovoltaic system model is constructed using the physical photovoltaic output model, and the improved CEF model is constructed;

[0051] The calculation module adds the net load of the enterprise and the DOS corresponding to each device to the constructed DOS estimation model, uses the improved Informer algorithm to train and predict the DOS of the device, and calculates the direct carbon emissions of the enterprise;

[0052] Add weather data to the constructed PV system model, use the physical PV output model to approximate the PV output, add PV equivalent output data to the constructed improved CEF model, and calculate the indirect carbon emissions of the enterprise;

[0053] By calculating direct carbon emissions and indirect carbon emissions, the total carbon emissions of the enterprise load are obtained;

[0054] The output module displays the calculated total carbon emissions of the enterprise load to the user in the form of charts, tables or reports.

[0055] The present invention has the following beneficial effects compared with the prior art:

[0056] 1. The enterprise load carbon emission detection method based on the photovoltaic equivalent output estimation model provided by the present invention analyzes the enterprise's direct carbon emissions by establishing an enterprise load equipment operation status model; models photovoltaics and adopts an equivalent output estimation model to reduce its impact on the accounting results; calculates the enterprise's indirect carbon emissions by constructing an improved carbon emission flow model; calculates the total carbon emissions including direct and indirect carbon emissions by establishing a comprehensive carbon emission calculation model; and realizes the functions of calculating the carbon emissions of industrial production and reducing the impact of photovoltaic access to the grid on carbon accounting;

[0057] 2. Compared with the existing corporate carbon emissions accounting methods based on financial data analysis, the present invention has lower reliance on sensitive data and is applicable to carbon emissions calculations in more scenarios and for more types of companies;

[0058] 3. The present invention fully considers the access problem of photovoltaic power generation system in practical application and the corresponding meter measurement problem, and reduces the impact of photovoltaic power generation on the carbon emission accounting results;

[0059] Fourth, the present invention calculates the carbon emissions in real time, rather than measuring the total carbon emissions for a period of time after production, which can effectively enhance the accuracy of the carbon emissions calculation results. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] The present invention will be further described below in conjunction with the accompanying drawings:

[0061] Figure 1 is a framework structure diagram of a DOS estimation model used in an embodiment of the present invention;

[0062] Figure 2 is a flow chart of the steps of photovoltaic system parameter estimation in an embodiment of the present invention;

[0063] Figure 3 It is a flow chart of the steps of detecting the carbon emission of enterprise load in the present invention. DETAILED DESCRIPTION

[0064] The present invention provides a method for detecting carbon emissions of enterprise loads based on a photovoltaic equivalent output estimation model, which is used to calculate the carbon emissions of industrial production and reduce the impact of photovoltaic access to the power grid on carbon accounting. Specifically, the direct carbon emissions of the enterprise are analyzed by establishing an enterprise load equipment operation status model, photovoltaic modeling is performed, and an equivalent output estimation model is adopted to reduce its impact on the accounting results. By constructing an improved carbon emission flow model, the indirect carbon emissions of the enterprise are calculated, and a comprehensive carbon emission calculation model is established to calculate the total carbon emissions including direct and indirect carbon emissions, and finally the purpose of calculating the carbon emissions of industrial production is achieved. The present invention has low dependence on sensitive data, is applicable to more scenarios and types of enterprises, and fully considers the access problem of photovoltaic power generation systems in actual applications and the corresponding meter measurement problem, reduces the impact of photovoltaic power generation on the carbon emissions accounting results, and the present invention calculates carbon emissions as real-time results, which enhances the accuracy of carbon emissions accounting results.

[0065] like Figure 3 As shown, the steps of detecting the carbon emissions of enterprise loads in the present invention include:

[0066] S1: Initialization, input relevant parameters;

[0067] S2: Establish a DOS (device operation state) estimation model, analyze the direct carbon emissions of the enterprise, and introduce and improve the Informer algorithm to estimate DOS;

[0068] Since each device has multiple operating states, DOS estimation is a multi-class NILM problem. The present invention introduces and improves the Informer algorithm to estimate DOS. The Informer algorithm improves the inference speed of long sequence prediction for existing deep learning methods through the probabilistic sparse ProbSparse self-attention mechanism, self-attention extraction and generative decoder. Since accurate DOS estimation requires high-frequency and long-term data, the Informer algorithm is suitable for such long-sequence time series prediction problems.

[0069] like Figure 1 As shown in Figure 1, the framework of the DOS estimation model includes three parts: data preparation, model training, and DOS estimation. The data includes the net load of the enterprise and the DOS corresponding to each device. First, in the data preparation stage, the values ​​of the net load data are scaled to between 0 and 1 by using min-max normalization.

[0070] Among them, the net load P NL,t is the actual load P L,t With photovoltaic output P PV,t The difference is expressed as:

[0071] P NL,t =P L,t -P PV,t ;

[0072] For a device, there are training load data X, test load data X' and predicted load data X'', which are K1, K2 and K3 dimensional vectors respectively; corresponding to X, X' and X'', the training, test and predicted DOS data are Y, Y' and Y'' respectively, which are category labels, converted into binary (only 0 and 1) matrices with dimensions of m×K1, m×K2 and m×K3 respectively.

[0073] Secondly, when training the model, the improved Informer algorithm is used. The initial Informer model is modified into a multi-category prediction model. There are four places in the model that need to be modified: the target data is changed to the category label of DOS, the CrossEntropyLoss function is selected as the loss function, the ReLU function is set as the activation function, and a dense layer is added at the end of the model. In addition, the structure and other settings of the improved algorithm are the same as the initial Informer algorithm. Finally, substitute X'' into the trained model to obtain Y'', which represents the estimated value of DOS.

[0074] S3: Modeling the photovoltaic system, mainly using a general physical photovoltaic model to approximate the photovoltaic output, in order to reduce the impact of the modeling process on the accounting results;

[0075] Photovoltaic solar panels are installed at different locations with different capacities, tilt angles and inclinations. It is difficult to measure the output of each photovoltaic solar panel without sufficient instruments. Since the tilt angles and latitudes of photovoltaic solar panels in adjacent areas are similar, the present invention sets the photovoltaic panels as a virtual equivalent photovoltaic system with a fixed tilt photovoltaic array.

[0076] The present invention uses a general physical photovoltaic model to approximate photovoltaic output. Its input includes weather data, solar irradiance data and photovoltaic system technical parameters. The weather data can be collected from nature. The solar irradiance data is the sum of direct reflection horizontal irradiance, diffuse reflection horizontal irradiance and direct reflection horizontal irradiance.

[0077] The technical parameters of the photovoltaic system that need to be estimated include: the capacity C of the decision variable photovoltaic power station, the tilt angle β and the azimuth angle γ of the photovoltaic panel; the photovoltaic output is approximately expressed as:

[0078] ;

[0079] In the formula, I PV,t is the solar radiation received by the photovoltaic panel at time t; μ is the temperature coefficient, which is equal to 0.5% / ℃;

[0080] T PV,t It represents the temperature of the photovoltaic module at time t, which is related to the cumulative effect of photovoltaic cell materials, outdoor ambient air temperature and solar radiation. PV,t It is approximately expressed by the following function:

[0081] ;

[0082] Where, T A,t is the outdoor ambient temperature at time t; N oct Indicates the rated operating unit temperature;

[0083] I PV,t It is the sum of direct reflection horizontal irradiance, diffuse reflection horizontal irradiance and direct reflection horizontal irradiance at time t, expressed as:

[0084] ;

[0085] Where: I 0,t is the direct normal irradiance at time t; τ b,t is the atmospheric transparency coefficient at time t; I b,t ,I h,t Respectively represent the direct and diffuse horizontal irradiance at time t; ρ t is the surface albedo at time t.

[0086] In order to estimate the output power of the photovoltaic system, the present invention models the parameter estimation as a nonlinear optimization problem, with decision variables C, β and γ, and the objective function is to minimize the sum of DOS estimation errors. Since the photovoltaic output is invisible, which will affect the DOS estimation result, the photovoltaic output is indirectly estimated through the DOS estimation accuracy, and the objective function is:

[0087] ;

[0088] Where: D is the total number of devices; d is the device index; η d represents the DOS estimation accuracy of device d; K is the total number of data samples; binary variable σ d,t is the accuracy of the estimated DOS at time t, which is a binary variable that is equal to 1 if the estimated DOS value of device d at time t is correct, otherwise it is equal to 0; d,t 、o' d,t Denote the actual and estimated DOS of device d at time t, respectively. Different distributions of PV have different estimated output accuracies, which leads to different accuracies of estimated DOS, thus affecting the accuracy of direct carbon emissions and the proposed method. Figure 2 As shown, the parameter estimation process for each PV system is demonstrated.

[0089] S4: Construct an improved carbon emission flow model and introduce the CEF (carbon emission flow) model to calculate the indirect carbon emissions of enterprises;

[0090] Indirect carbon emissions from industrial enterprises mainly come from electricity consumption. Electricity carbon emissions are determined by the net load of the node holding company in the power system and the carbon emission intensity (CEI) of electricity. Generally speaking, indirect carbon emissions are time-varying because both the load and the node power CEI are constantly changing. Net load can be measured by smart meters, but node power CEI is difficult to obtain. Before calculating the node power CEI, the operating status of the power system must be known.

[0091] In order to calculate the real-time node CEI, the present invention introduces the CEF model. However, if enterprises install photovoltaics, they may become consumers at some point. In this case, the original CEF model will no longer be applicable to calculate indirect carbon emissions. Therefore, the present invention proposes an improved CEF model that considers prosumers. The model regards CEF as a virtual flow together with the power flow. For an n-node power system at time t, the expression of node power CEI can be derived as:

[0092] ;

[0093] In the formula, e N,t For the momentt Column vector of node power CEI; R C,t For the moment t Column vector of the released carbon emission flow; e N,t and R C,t are all N-dimensional column vectors, P NB,t is the discriminant matrix calculated at time tCEF; P B,out,t is the branch output power distribution matrix at time t; P NB,t and P B,out,t are all N-dimensional square matrices; t is the unit (N+G+P)-dimensional row vector; P G,t is the power injection distribution matrix at time t, which is an N×(G+P)-dimensional column vector; E G,t For the moment t The column vector of power CEI is a (G+P)-dimensional column vector; P gen,t is the power injection distribution matrix of the generator at time t, which is an N×G dimensional matrix; P pro,t is the power injection distribution matrix of the prosumer at time t, which is an N×P dimensional matrix; E gen,t is the column vector of producer CEI at time t, which is a G-dimensional column vector; E pro,t is the column vector of the CEI of the prosumer at time t, and is a P-dimensional column vector. "diag{}" is a diagonal matrix operator.

[0094] Furthermore, the calculation formula for node electricity carbon emissions is:

[0095] ;

[0096] In the formula, is the column vector of carbon emissions at the node at time t; P NL,t is the column vector of node net load at time t; and P NL,t All are N-dimensional column vectors. In addition, the present invention does not consider the power loss of the transmission line.

[0097] The difference between the improved CEF model and the original model is mainly in P G,t , E G,t , P B,out,t and P NL,t On the one hand, for a prosumer at node k, when it acts as a producer at time t, the power of the prosumer needs to be injected into the distribution matrix P pro,t and the column vector E of the prosumer CEI pro,t The kth value in is changed from 0 to the actual power value and from 0 to the CEI value of the prosumer. Since the CEI of photovoltaic is 0, the present invention changes E pro,t The kth value in is set to 0. At the same time, PNL,t The load value of node k needs to be set to 0. In addition, considering this prosumer, P needs to be recalculated. B,out,t On the other hand, when this prosumer acts as a consumer at time t, P pro,t and E pro,t All are zero vectors, and the improved model is the same as the original model. The improved carbon emission flow model proposed by the present invention can accurately calculate indirect CEA.

[0098] S5: Establish a comprehensive carbon emissions calculation model to calculate the total carbon emissions including direct and indirect carbon emissions.

[0099] In the present invention, the CEA scope of an enterprise includes direct carbon emissions and indirect carbon emissions. The total carbon emissions of an enterprise at node k in the power system are expressed as:

[0100] ;

[0101] In the formula, E k is the total carbon emission of node k; E k,scope1 is the direct carbon emission of node k; E k,scope2 is the indirect carbon emission of node k; K is the total number of data samples; D is the total number of devices; E device,d,t is the direct carbon emission of device d at time t; is the carbon emission of electricity at node k at time t; e d,t is the CEI of device d at time t; o' d,t is the estimated DOS of device d; constant value e d1 To e dm is the CEI of device d in different working states.

[0102] The order of execution of each step in the above embodiment does not mean the order of execution. The order of execution of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiment of the present application.

[0103] In order to verify the effectiveness of the proposed enterprise load carbon emission accounting method based on the photovoltaic equivalent output estimation model, the present invention uses the following three different enterprise load carbon emission accounting methods for comparison:

[0104] Method 1: Using CNN-BLSTM algorithm to estimate DOS without considering PV output estimation;

[0105] Method 2: Using the improved Informer algorithm proposed in the present invention to estimate DOS without considering the photovoltaic output estimation;

[0106] Method 3: The improved Informer algorithm proposed in the present invention is used to estimate DOS, and the photovoltaic output estimation model proposed in the present invention is used.

[0107]

[0108] Table 1 Comparison of enterprise load carbon emission accounting results under different methods

[0109] Table 1 compares the enterprise load carbon emission accounting results under the three methods provided by the present invention. As the photovoltaic output ratio increases, the accuracy of DOS estimation gradually decreases. The accuracy of method three proposed by the present invention is reduced by about 10%, which is the least reduced in accuracy compared to the other two traditional methods. When the photovoltaic output ratio is 100%, the DOS estimation accuracy of method one and method two drops to 41.32% and 71.67% respectively. At the same time, the estimation results of method two are more accurate than method one in the final enterprise load carbon emission accounting results under different photovoltaic output ratios, which shows that the improved Informer algorithm of the present invention is superior to the traditional CNN-BLSTM algorithm.

[0110] From the enterprise load carbon emission accounting results of method 2, it can be seen that if the photovoltaic output ratio increases, DOS cannot be accurately estimated. For example, when the photovoltaic output ratio increases to 30%, the DOS estimation accuracy drops to 81.21%, indicating that accurate load data is crucial to the enterprise load carbon emission accounting results. Since photovoltaics hide part of the actual load, the accuracy of the enterprise load carbon emission accounting results will be greatly affected.

[0111] From the enterprise load carbon emission accounting results of method 3, it can be seen that the accuracy under different photovoltaic output ratios is about 90%, which is significantly higher than the accuracy based on method 1 and method 2. This shows that the enterprise load carbon emission accounting method based on the photovoltaic equivalent output estimation model proposed by the present invention can obtain accurate calculation results.

[0112] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for detecting carbon emissions from enterprise loads, characterized in that: The detection steps include the following: S1: Initialization, inputting parameters related to enterprise load carbon emissions; S2: Introduce and improve the Informer algorithm, establish a DOS estimation model for the operating status of enterprise load equipment, and analyze the direct carbon emissions of enterprises; S3: Use the general physical photovoltaic model to approximate the photovoltaic output effect and establish a photovoltaic system model; The specific method of using the general physical photovoltaic model to approximate the photovoltaic output effect in S3 is: The photovoltaic panels are set as a virtual equivalent photovoltaic system with a fixed tilt photovoltaic array; Inputs include weather data, solar irradiance data, and PV system technical parameters; The PV output is approximated using a general physical PV model, and the PV system technical parameters that need to be estimated include: the capacity C of the PV power plant, the tilt angle β, and the orientation γ of the PV panels, which are decision variables; Let D be the total number of devices, η d is the DOS estimation accuracy of device d, K is the total number of data samples; binary variable σ d,t is the accuracy of the estimated DOS at time t. If the estimated DOS value of device d at time t is correct, then σ d,t is equal to 1, otherwise σ d,t Equal to 0; o d,t ,o' d,t are the actual and estimated DOS of device d at time t respectively; the PV output is indirectly estimated through the DOS estimation accuracy, and the objective function is to minimize the sum of DOS estimation errors, expressed as: ； Let I PV,t is the solar radiation received by the photovoltaic panel at time t, T A,t is the outdoor ambient temperature at time t, N oct is the rated working unit temperature, then the temperature of the photovoltaic module at time t is T PV,t Approximate expression is: ； Assuming μ is the temperature coefficient, the photovoltaic output P PV,t Approximate expression is: ； The specific method of establishing the photovoltaic system model in S3 is: Let I 0,t is the direct normal irradiance at time t; τ b,t is the atmospheric transparency coefficient at time t; I b,t ,I h,t They represent the direct reflection horizontal irradiance and diffuse reflection horizontal irradiance at time t respectively; ρ t is the surface albedo at time t, α is the height of the sun, A is the direction of the sun, then the solar radiation I received by the photovoltaic panel at time t PV,t It is the sum of direct reflection horizontal irradiance, diffuse reflection horizontal irradiance and direct reflection horizontal irradiance at time t, and the expression is: ； S4: Construct an improved carbon emission flow CEF model to calculate the indirect carbon emissions of enterprises; The specific method for constructing the improved carbon emission flow CEF model in S4 is: Calculate the real-time node CEI, assuming P NL,t is the node net load column vector at time t, e N,t is the column vector of the CEI of the node electricity at time t, then the carbon emission vector of the node electricity at time t is The expression is: ； in, and P NL,t They are all N-dimensional column vectors, and Δt is the time interval; S5: Establish a comprehensive carbon emissions calculation model to calculate the total carbon emissions including direct carbon emissions and indirect carbon emissions.

2. The method for detecting carbon emissions from enterprise load according to claim 1, characterized in that: The enterprise load equipment operation status DOS estimation model established in step S2 includes data preparation, model training and DOS estimation. The data includes the net load of the enterprise and the DOS corresponding to each device. The specific modeling method is: S21: In the data preparation stage, min-max normalization is used to convert the net load P NL,t The value is scaled to between 0 and 1, where the net load P NL,t is the actual load P L,t With photovoltaic output P PV,t The difference is expressed as: P NL,t =P L,t -P PV,t ; For a device, there are training load data X, test load data X' and predicted load data X'', which are K1, K2 and K3 dimensional vectors respectively; corresponding to the data X, X' and X'', the training, test and predicted DOS data are Y, Y' and Y'' respectively, which are used as category labels and converted into binary matrices of dimensions m×K1, m×K2 and m×K3 respectively, where m represents the number of operating states of the device; S22: Use the improved Informer algorithm to estimate DOS when training the model, and modify the initial Informer model into a multi-class prediction model; S23: Substitute the predicted load data X'' into the trained model to obtain the predicted DOS data Y''.

3. The method for detecting carbon emissions from enterprise load according to claim 1, characterized in that: The Informer algorithm introduced and improved in step S2 includes ProbSparse self-attention mechanism, self-attention extraction and generative decoder; The improvements to the Informer model include: changing the target data to the category label of DOS, selecting the CrossEntropyLoss function as the loss function, setting the ReLU function as the activation function, and adding a dense layer at the end of the model.

4. The method for detecting carbon emissions from enterprise load according to claim 1, characterized in that: The specific method for calculating the indirect carbon emissions of the enterprise in step S4 is: Assume R C,t is the column vector of carbon emission flow released at time t, e N,t and R C,t are all N-dimensional column vectors; P NB,t Calculate the discriminant matrix for CEF at time t, P B,out,t is the branch output power distribution matrix at time t, P NB,t and P B,out,t They are all N-dimensional square matrices; ζ t is the unit (N+G+P)-dimensional row vector; P G,t is the power injection distribution matrix at time t, which is an N×(G+P)-dimensional column vector; E G,t is the column vector of the power CEI at time t, which is a (G+P)-dimensional column vector; P gen,t is the power injection distribution matrix of the generator at time t, which is an N×G dimensional matrix; P pro,t is the power injection distribution matrix of the prosumer at time t, which is an N×P dimensional matrix; E gen,t is the column vector of producer CEI at time t, which is a G-dimensional column vector; E pro,t is the column vector of the CEI of the prosumer at time t, which is a P-dimensional column vector; "diag{}" is a diagonal matrix operator; considering CEF as a virtual flow together with the power flow, for the n-node power system at time t, the expression of the real-time node CEI is: 。 5. The method for detecting carbon emissions of an enterprise load according to claim 1, characterized in that: The specific method for establishing the comprehensive carbon emission calculation model and calculating the total carbon emission in step S5 is: Assume E k is the total carbon emission of node k; E k,scope1 is the direct carbon emission of node k; E k,scope2 is the indirect carbon emission of node k; K is the total number of data samples; D is the total number of devices; E device,d,t is the direct carbon emission of device d at time t; is the carbon emission of electricity at node k at time t; e d,t is the CEI of device d at time t; o' d,t is the estimated DOS of device d at time t; constant value e d1 To e dm is the CEI of equipment d under different working conditions; then the expression of the total carbon emissions of the enterprise at node k is: 。 6. The detection system used in the enterprise load carbon emission detection method according to claim 1 is characterized by: The detection device includes the following: The main server is responsible for the management and scheduling of the overall carbon emissions calculation task and has data processing and storage functions; Computing nodes: Multiple computing nodes are connected to the main server through the network to form a distributed computing cluster. Each computing node is responsible for executing specific carbon emission calculation tasks. The control terminal is used for user interaction and has the functions of inputting calculation parameters and viewing calculation results. The control terminal is connected to the main server through a local area network or the Internet; The data acquisition terminal is responsible for collecting real-time load data and weather data of the enterprise power grid and sending the data to the main server through the data transmission channel.

7. The detection system used in the enterprise load carbon emission detection method according to claim 6 is characterized by: The main server includes: The data preprocessing module is used to receive the enterprise net load, DOS data and meteorological data from the data acquisition terminal, and perform data cleaning, format conversion, abnormal value detection and processing; A construction module is used to obtain the carbon emission data of enterprises and establish a comprehensive carbon emission calculation model that considers direct carbon emissions and indirect carbon emissions. Specifically, the DOS estimation model is constructed using the improved Informer algorithm, the photovoltaic system model is constructed using the physical photovoltaic output model, and the improved CEF model is constructed; The calculation module adds the net load of the enterprise and the DOS corresponding to each device to the constructed DOS estimation model, uses the improved Informer algorithm to train and predict the DOS of the device, and calculates the direct carbon emissions of the enterprise; Add weather data to the constructed PV system model, use the physical PV output model to approximate the PV output, add PV equivalent output data to the constructed improved CEF model, and calculate the indirect carbon emissions of the enterprise; By calculating direct carbon emissions and indirect carbon emissions, the total carbon emissions of the enterprise load are obtained; The output module displays the calculated total carbon emissions of the enterprise load to the user in the form of charts, tables or reports.

Images