A power grid node carbon emission distribution probability tracking method and system

Abstract

The application discloses a power grid node carbon emission distribution probability tracking method. A cluster-adaptive power flow distribution matrix is constructed, a linear mapping of power generation power and node power flow is established, and a clustering algorithm is used to divide the generators into several clusters according to carbon measurement heterogeneity; secondly, the stable distribution basic parameters of each generator cluster are calibrated differently, a cluster-level time-varying influence factor set is constructed, a time-varying parameter model is trained, and a special modeling of extreme carbon measurement error is combined with extreme value distribution to construct a mixed distribution model; finally, the distribution parameters of each cluster to the carbon flow rate of each load node are derived, and the extreme error correction parameter transmission and the cluster contribution weight fusion are combined to obtain the final mixed distribution model of the load node carbon flow rate. The application accurately depicts the full-dimensional uncertainty of carbon measurement, realizes the probabilistic transmission of the carbon emission of the power generation side to the load side, and supports the high space-time resolution probability tracking of the node carbon emission under the conditions of full voltage level and multiple scenes.

Description

Technical Field

[0001] This invention belongs to the field of power system carbon emission metering and management technology, specifically involving a probabilistic tracking method for the distribution of carbon emissions at power grid nodes based on heterogeneous clustering and time-varying extreme error correction. It is applicable to high spatiotemporal resolution probabilistic tracking of carbon emissions at nodes of all voltage levels in a provincial power grid. Background Technology

[0002] Driven by the "dual carbon" goals, accurate metering and node-level distribution tracking of carbon emissions in the power system have become core technical support for low-carbon grid dispatching and carbon emission responsibility sharing. In particular, provincial power grids need to take into account the power flow exchange of inter-provincial interconnections, the output characteristics of new energy power plants outside the province, and the spatiotemporal transfer patterns of carbon emissions at nodes of all voltage levels, which significantly increases the requirements for high spatiotemporal resolution and accuracy in carbon emission distribution tracking.

[0003] However, existing carbon emission tracking methods for power systems have the following major drawbacks: First, there is a lack of probabilistic modeling for the uncertainties in carbon metering on the power generation side. Existing methods mostly focus on deterministic calculations of physical currents, without fully considering the uncertainties caused by equipment accuracy, changes in operating conditions, and extreme anomalies during the carbon metering process on the power generation side. This makes it difficult to quantify the probabilistic distribution characteristics of carbon emissions during the transfer process from the power generation side to the load side, and thus cannot provide probabilistic data support for carbon emission risk assessment and accurate responsibility allocation.

[0004] Second, the existing methods do not perform clustered modeling for generator heterogeneity. The types of generators within a provincial power grid are diverse, and the carbon metering error characteristics, operation and maintenance levels, and degree of influence from operating conditions differ significantly among coal-fired, gas-fired, and new energy units. Existing methods only employ a uniform error processing approach and do not perform clustered modeling for generator heterogeneity, resulting in significant deviations between carbon emission tracking results and actual operating conditions.

[0005] Third, static modeling methods cannot capture the time-varying characteristics of carbon measurement errors. Carbon measurement errors change dynamically with factors such as system load rate, ambient temperature, and equipment operating time. Traditional static modeling methods cannot capture the time-varying characteristics of these errors, further reducing the accuracy of tracking node carbon emission distribution.

[0006] Fourth, the existing technology has not established a strong correlation between "implicit" carbon emissions and power grid physical flow, and has not considered the impact of extreme carbon metering errors. The existing technology has not established a strong correlation between carbon emissions and actual physical flow, and has not considered the impact of extreme carbon metering errors on carbon emission distribution. Conventional deterministic tracking models cannot output the probability distribution range of node carbon emissions, making it difficult to meet the full-dimensional quantitative requirements of carbon emission distribution for power grid low-carbon planning and dispatching decisions. Summary of the Invention

[0007] The technical solution of this invention is used to solve the problem of probabilistic transmission of carbon emissions from the power generation side to the load side and accurate tracking of node-level distribution.

[0008] The present invention solves the above-mentioned technical problems through the following technical means: A method for probabilistically tracking the carbon emission distribution of power grid nodes includes: Step 1. Construct a clustered power flow distribution matrix, establish a linear mapping between power generation and node flow power, and use a clustering algorithm to divide the generators into several clusters based on carbon metering heterogeneity; Step 2. For each generator cluster, calibrate the basic parameters of the Alpha stable distribution in a differentiated manner, construct a cluster-level time-varying influencing factor set, train the time-varying parameter model to dynamically update the distribution parameters, and combine the Gumbel extreme value distribution to specifically model extreme carbon measurement errors, and construct a time-varying Alpha stable and Gumbel extreme value hybrid distribution model. Step 3. Based on the linear combination characteristics of the Alpha stable distribution, the distribution parameters of the contribution of each cluster to the carbon flow rate of each load node are analytically derived, and the final mixed distribution model of the carbon flow rate of the load node is obtained by passing the extreme error correction parameters and integrating them with the inter-cluster contribution weights.

[0009] Furthermore, step 1 specifically involves: constructing a clustered adaptive power flow distribution matrix and calculating the power flowing through the nodes; defining the power balance relationship for each node in the power system according to the power conservation law, and reorganizing the power balance relationship of each node into a matrix equation. Construct the power flow distribution matrix Finding its inverse yields ,pass Calculate the power vector flowing through the nodes Collect generator operating characteristic data and historical carbon metering error data, construct quantitative clustering feature vectors, and use the K-means clustering algorithm to divide all generators in the system into groups. Clusters, generator power vector Split into clusters .

[0010] Furthermore, step 2 specifically involves: clustered time-varying Alpha stable distribution modeling, and constructing a hybrid distribution model for each generator cluster. Cluster-level stability parameters are differentiated using the maximum likelihood estimation method. And calibrate the basic skewness parameters of each generator in the cluster. Basic Scale Parameters Basic position parameters For each cluster, a cluster-level time-varying influence factor set is constructed, and time-varying parameter models for skewness, scale, and location are trained, with the time-varying parameters updated in real time. , , A time-varying alpha stable distribution of carbon emission intensity for each generator was established; the cluster-level extreme carbon measurement error threshold was determined using the 3σ principle combined with the 99th percentile method. The extreme error interval is modeled using the Gumbel extreme value distribution to construct a mixed distribution model, and the effectiveness is verified by the KS test.

[0011] Furthermore, step 3 specifically involves: clustered linear transfer and inter-cluster fusion, deriving the carbon flow rate mixing distribution parameters of the load nodes: for each generator cluster According to the inverse matrix of the power flow distribution matrix Calculate the linear combination coefficient of carbon flow rate of each generator in the cluster to each load node. The distribution parameters of each cluster's contribution to the carbon flow rate at each load node are analytically derived using the linear combination property of the Alpha stable distribution. The extreme carbon flow rate error correction parameters for each cluster are obtained by weighted fusion according to the Gumbel extreme value distribution parameters. Finally, the weights of each cluster's contribution to the carbon flow rate at each load node are calculated. The distribution parameters of each cluster and the extreme error correction parameters are integrated according to the contribution weight to obtain the final mixed distribution model of carbon flow rate of each load node.

[0012] Furthermore, the power flow distribution matrix mentioned in step 1 The element is defined as:

[0013] Among them, This is the power vector flowing through the nodes, with diagonal elements set to 1 and off-diagonal elements representing upstream nodes. For downstream nodes The power contribution ratio of the elements is zero; the remaining elements are zero.

[0014] Furthermore, the construction of the quantitative clustering feature vector in step 1 includes: using carbon emission intensity as the target variable, and constructing a generator operation feature set based on unit thermal state characteristics, pressure and vacuum characteristics, steam and water and flue gas flow characteristics, load and energy consumption characteristics, and flue gas emission characteristics; further combining the central tendency, dispersion, skewness, and tail characteristics of historical time-series data of carbon metering errors, as well as the model specifications, range, accuracy level, sampling frequency, and operation and maintenance status of carbon metering equipment, to construct a quantitative clustering feature vector and perform normalization processing; determining the optimal number of clusters based on the silhouette coefficients corresponding to different candidate cluster numbers; and updating the quantitative clustering feature vector and re-clustering when carbon metering equipment is replaced, equipment parameters are adjusted, or operation and maintenance status changes.

[0015] Furthermore, the core factors of the cluster-level time-varying influence factor set mentioned in step 2 include: system load rate, generator real-time output fluctuation coefficient, carbon metering equipment operating time, and ambient temperature; for new energy unit clusters, the new energy output penetration rate is added as a time-varying influence factor; the time-varying influence factor sets of different clusters are set differently.

[0016] Furthermore, the time-varying parameter model mentioned in step 2 is determined from ARIMA, LSTM, and XGBoost models, wherein: the ARIMA model determines the model order according to the minimum AIC or BIC criterion; the LSTM and XGBoost models determine the model structure and parameters based on the prediction error of the validation set; the model with the smallest overall prediction error of the validation set is selected as the time-varying parameter model for the corresponding cluster; the Bootstrap repeated sampling method is used to evaluate the time-varying parameter model to obtain the predicted values ​​of each time-varying parameter and the corresponding confidence intervals, so as to quantify the uncertainty of the time-varying parameter estimation.

[0017] Furthermore, the personalized correction coefficient mentioned in step 2 The generator's basic parameters are determined based on the deviation between the cluster's generator operation and maintenance level and the cluster's average level; these parameters are then fine-tuned using personalized correction coefficients to obtain the final generator basic parameters: All generators within the cluster use cluster-level stability parameters. This satisfies the requirements of the linear combination theory of the stable Alpha distribution.

[0018] Furthermore, the construction rule for the hybrid distribution model described in step 2 is: within the normal error interval ( The time-varying Alpha stable distribution probability density function is adopted; in the extreme error interval ( The probability density function of the Gumbel extreme value distribution is used instead; a probability smoothing coefficient is introduced at the threshold junction. This achieves a continuous transition between the two distribution segments, avoiding abrupt probability changes; where PDF represents the probability density function. The probability density function is a mixed distribution. For a mixed distribution model, For probability smoothing system, This represents the probability density function of the stable Alpha distribution. Let Gumbel be the probability density function of the extreme value distribution.

[0019] The present invention also provides a probability tracking system for carbon emission distribution at power grid nodes, comprising: Matrix construction module: Construct a cluster-adaptive power flow distribution matrix, establish a linear mapping between power generation and node flow power, and use a clustering algorithm to divide generators into several clusters based on carbon metering heterogeneity; The module for constructing a hybrid distribution model is as follows: Alpha stable distribution basic parameters are calibrated differently for each generator cluster, a cluster-level time-varying influencing factor set is constructed, a time-varying parameter model is trained to dynamically update the distribution parameters, and extreme carbon measurement errors are specifically modeled by combining Gumbel extreme value distribution, thus constructing a time-varying Alpha stable and Gumbel extreme value hybrid distribution model. Analytical derivation module: Based on the linear combination characteristics of the stable Alpha distribution, the distribution parameters of the contribution of each cluster to the carbon flow rate of each load node are analytically derived. By transferring the extreme error correction parameters and fusing them with the contribution weights between clusters, the final mixed distribution model of the carbon flow rate of the load nodes is obtained.

[0020] The advantages of this invention are: (1) Accurately characterize the uncertainty of carbon measurement across all dimensions. By combining K-means clustering modeling and time-varying parameter design with a mixture of Alpha stable distribution and Gumbel extreme value distribution, the system adapts to the heterogeneity of generator carbon measurement errors, captures the time-varying characteristics of errors and extreme scenarios, significantly improves the accuracy of carbon emission probability calculation, and overcomes the limitations of traditional methods in unifying error processing.

[0021] (2) Achieve refined probabilistic transfer of carbon tidal load sources. Construct a cluster-adaptive tidal distribution matrix, combine cluster linear transfer with inter-cluster weight fusion, and rely on the linear combination characteristics of the stable Alpha distribution to complete the precise allocation of carbon emissions at the node level, adapt to the calculation requirements of all voltage levels, and provide a probabilistic basis for the accurate division of carbon emission responsibilities.

[0022] (3) Achieve high spatiotemporal resolution carbon emission probability tracking. Based on the core methods of real-time time-varying modeling and linear mapping, a precise linear mapping between power generation and node flow power is established through the power flow distribution matrix. Combined with real-time operating data, the distribution parameters are dynamically updated to achieve high spatiotemporal resolution carbon emission probability tracking and support accurate calculation under power flow changes in multiple scenarios.

[0023] (4) Effectively address extreme carbon measurement error scenarios. The extreme error threshold is identified by combining the 3σ principle with the 99th percentile method. The Gumbel extreme value distribution is introduced to specifically model extreme carbon measurement errors. A smoothing coefficient is used to achieve a continuous transition between the normal error range and the extreme error range, ensuring the rationality and robustness of the mixed distribution model. Attached Figure Description

[0024] Figure 1 This is a schematic diagram of the overall process of the method in the embodiments of the present invention; Figure 2 This is a schematic diagram of the clustered time-varying hybrid distribution modeling process in step 2 of this embodiment of the invention; Figure 3This is a schematic diagram comparing the theoretical carbon flow rate distribution of a typical load node with Monte Carlo simulation verification in an embodiment of the present invention (2-machine 4-node application embodiment). Figure 4 This is a schematic diagram comparing the theoretical calculation results of the average carbon flow rate of each load node in the IEEE 30-node system with the Monte Carlo simulation in an embodiment of the present invention. Detailed Implementation

[0025] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below in conjunction with the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0026] The technical solution of the present invention will be described in detail below with reference to the accompanying drawings and embodiments. In this specification, subscripts are used to indicate: G represents a generator, L represents a load, u represents an upstream-downstream relationship, and C represents the total cluster size.

[0027] This embodiment describes a method for probabilistic tracking of carbon emission distribution at power grid nodes, such as... Figure 1 As shown, it includes the following steps: Step 1: Construct the power flow distribution matrix for cluster adaptation and calculate the power flowing through the nodes. This step forms the basis for carbon flow tracking. Its core is to establish a linear mapping relationship between power generation and the power flowing through each node of the system, while also adapting to the subsequent clustering modeling requirements. The specific steps are as follows: 1.1 Define the node power balance relationship For any node in the power system ( , (where the total number of nodes in the power system is 1), the power balance relationship is defined according to the power conservation law:

[0028] in, For nodes The power flowing through (the total power flowing into this node). For nodes The set of upstream nodes, For nodes Flow to Node Inflow node in the lossy branch power, For access nodes The generator's injected power.

[0029] 1.2 Constructing the power flow distribution matrix The node power balance relationship is organized into a matrix form:

[0030] in The power vector flowing through the node. Generator power injection vector; power flow distribution matrix The elements are defined according to the following rules:

[0031] Its matrix has diagonal elements of 1, satisfying the power balance characteristic; the off-diagonal elements are upstream nodes. For downstream nodes The power contribution ratio is uniquely determined by the grid topology and power flow distribution.

[0032] 1.3 Calculate the power vector flowing through the nodes Inverting the power flow distribution matrix yields By using linear transformation, the power flowing through each node from the generator is accurately distributed. The formula for calculating the power vector flowing through each node is as follows:

[0033] 1.4 Generator Clustering and Power Vector Splitting For generators with direct carbon emissions within the system, carbon emission intensity is used as the target variable, and time is used as the time series identifier. The remaining monitoring data are divided into unit thermal state characteristics, pressure and vacuum characteristics, steam, water, and flue gas flow characteristics, load and energy consumption characteristics, and flue gas emission characteristics, forming a generator operation feature set. Furthermore, by combining the central tendency, dispersion, skewness, and tail characteristics of historical time series data on carbon metering errors for each generator, as well as the model, specifications, range, accuracy level, sampling frequency, and maintenance status of the carbon metering equipment, a quantitative clustering feature vector is constructed. This quantitative clustering feature vector is normalized, and the K-means clustering algorithm is used to divide all generators within the system into groups. There are several clusters, and the optimal number of clusters is determined based on the contour coefficients corresponding to different candidate cluster numbers, thus dividing the generators in the system into ( ). Generators within the same cluster exhibit highly similar carbon metering uncertainty characteristics. When carbon metering equipment is replaced, equipment parameters are adjusted, or the operational status changes, the quantized clustering feature vector is updated, and clustering is performed again. (The last sentence appears to be incomplete and unrelated to the preceding text.) quantity Split into clusters This prepares for subsequent cluster modeling and carbon flow rate derivation.

[0034] The entire output of step 1 is: power flow distribution matrix. Inverse matrix Power vector flowing through each node Generator power vector split by cluster The generator cluster division results and the labels of each generator cluster.

[0035] Step 2: As Figure 2 As shown, clustered time-varying Alpha stable distribution modeling is used to construct a hybrid distribution model. This step addresses the uncertainty in source-side carbon metric analysis. Building upon the Alpha stable distribution model, it incorporates three practical characteristics: clustering, time-varying, and mixed distribution. This provides a more accurate source-side basis for modeling clustered, time-varying, and mixed distributions, enabling more accurate source-side modeling of carbon flow rate uncertainty propagation in step 3. Specific steps are as follows: Note: In this step, the subscripts of the symbols indicate the total number of generator clusters, c represents the c-th generator cluster, and 0 represents the basic state. is the cluster number, and th represents the threshold.

[0036] 2.1 Calibration of basic parameters for clustered alpha stable distribution (clustered modeling) 2.1.1 Cluster-level stability parameters Differential calibration For each generator cluster ( The cluster collected continuous carbon measurement data (measured values ​​+ standard true values) from all generators within the cluster for at least one complete operating cycle. Using the maximum likelihood estimation method, the cluster-level stability parameters were calibrated separately. ( ), different clusters Differentiated settings are possible, but all generators within the same cluster can use the same settings. This satisfies the requirements of the linear combination theory of the stable Alpha distribution.

[0037] 2.1.2 Calibration of Basic Parameters of In-Cluster Generators Cluster Each generator inside (recorded as) , , (Based on the number of generators within the cluster) and its historical carbon metering data, the basic skewness parameters of a single generator within the cluster are calibrated. Basic Scale Parameters Basic position parameters ,in , , , Characterizing skewness Describing the amplitude of fluctuations Describe the nominal level.

[0038] 2.1.3 Fine-tuning of individual parameters within a cluster According to the cluster generator To account for individual characteristics (such as the deviation between the operational level and the cluster average), a personalized correction coefficient is introduced. By fine-tuning the basic parameters, we can obtain... The final basic parameters:

[0039] Guarantee cluster It aims for uniformity while also taking into account the individual differences of each generator.

[0040] 2.2 Clustered Time-Varying Parameter Modeling (Time-Varying Modeling) 2.2.1 Determining the Clustered Time-Varying Influence Factor Set For each cluster Based on their cluster characteristics (coal-fired / gas-fired / new energy), a cluster-level time-varying influencing factor set is constructed. m represents the number of time-varying influencing factors at the cluster level. Core factors include: system load rate, generator real-time output fluctuation coefficient, carbon metering equipment operating time, and ambient temperature. For new energy unit clusters, the new energy output penetration rate is added. The influencing factor sets of different clusters can be set differently to fit the actual operating conditions.

[0041] 2.2.2 Training a cluster-level time-varying parameter model Time-series monitoring data (real-time monitoring values ​​of time-varying influencing factors and measured values ​​of generator carbon metering errors within the corresponding time periods) are collected from each cluster. The data is divided according to a preset sliding time window, and based on the measured carbon metering error values ​​within each time window, the skewness parameter, scale parameter, and location parameter for the corresponding time period are labeled to form a cluster-level time-varying parameter label sequence. Using the cluster-level basic parameters labeled in step 2.1 as a benchmark, and based on the time-varying influencing factor set... Using the time-series variation of carbon metrology error as the input feature and the time-series parameter label as the output label, the correlation between each time-varying influencing factor and the time-varying parameter label is calculated. The response curves of the skewness parameter, scale parameter, and location parameter when each time-varying influencing factor changes are fitted. Time-varying influencing factors with a correlation degree reaching a preset threshold are selected, and skewness time-varying models for each cluster are trained respectively. Scale-time-varying model Location time-varying model .

[0042] The time-varying model is determined from ARIMA, LSTM, and XGBoost models. For ARIMA, the model order is determined based on the minimum AIC or BIC criterion. For LSTM and XGBoost models, the model structure and parameters are determined based on the prediction error on the validation set. The model with the smallest overall prediction error on the validation set is selected as the time-varying parameter model for the corresponding cluster. The Bootstrap repeated sampling method is used to evaluate the time-varying parameter model, obtaining the predicted values ​​of each time-varying parameter and their corresponding confidence intervals to quantify the uncertainty of the time-varying parameter estimation.

[0043] 2.2.3 Real-time Time-varying Parameter Update Collect real-time operating condition monitoring data of the power system and extract time-varying influence factor sets for each cluster. The real-time values ​​are substituted into the corresponding skewness time-varying model, scale time-varying model, and location time-varying model to calculate the time. Lower cluster Cluster-level time-varying parameters , , and its confidence interval; combined with the individual correction coefficients of individual generators within the cluster. Get the time Down Real-time time-varying parameters:

[0044] Cluster level If it remains unchanged, it satisfies the requirements of linear combination.

[0045] 2.2.4 Determination of the Time-Varying Alpha Stable Distribution time Below, cluster Internal generator carbon emission intensity It follows a time-varying Alpha-stable distribution, denoted as:

[0046] 2.3 Construction of Hybrid Distribution Model (Hybrid Modeling) 2.3.1 Extreme Error Threshold Identification For each cluster Based on its historical carbon measurement error data, the cluster-level extreme carbon measurement error threshold was determined using the 3σ principle + 99% quantile method. Divide the regular error interval ( ) and extreme error range ( ),in This represents the relative error in carbon measurement. The extreme error range corresponds to extreme scenarios such as metering equipment failure and data anomalies.

[0047] 2.3.2 Calibration of Extreme Value Distribution of Extreme Errors Samples within the extreme error intervals of historical data from each cluster are extracted, and the extreme errors are modeled using the Gumbel extreme value distribution. The location parameters of the Gumbel distribution are then determined using the maximum likelihood estimation method. and scale parameters The extreme value distribution model of cluster-level extreme errors is obtained. ,in These are extreme error values.

[0048] 2.3.3 Construction of Hybrid Distribution For time generator below Construct a hybrid distribution model of "main distribution + local correction". PDF represents the probability density function. The probability density function is a mixture distribution, and the rules are as follows: (1) Conventional error range ( ): , The time-varying Alpha stable distribution probability density function obtained in step 2.2 is directly used to characterize the random error of conventional carbon metrology. (2) Extreme error range ( ): , The probability density function of the Gumbel extreme value distribution is used to replace the original Alpha stable distribution, thus accurately capturing extreme errors. (3) Threshold transition: Introduce a probability smoothing coefficient ,in This achieves a continuous transition between the main distribution and the modified distribution, avoids abrupt probability changes, and ensures the rationality of the distribution model, as detailed below: .

[0049] 2.3.4 Validation of Mixed Distribution The Kolmogorov-Smirnov test (KS test) was used to verify the goodness of fit between the mixed distribution model and the actual carbon metering error data of the generator. If the goodness of fit was ≥95%, the model was considered effective; if the goodness of fit was <95%, the extreme error thresholds were readjusted. Then calibrate the extreme value distribution parameters until the model meets the requirements.

[0050] The entire output of step 2 includes: generator cluster partitioning results and stability parameters of each cluster. Real-time time-varying hybrid distribution model of each generator Random variables of carbon emission intensity of each generator Cluster-level extreme error threshold And the Gumbel extreme value distribution parameters.

[0051] Step 3: Clustered linear transfer and inter-cluster fusion to derive the mixing distribution parameters of carbon flow rate at load nodes. This step is the core of carbon flow tracking. It achieves full-dimensional uncertainty quantification of carbon flow rate at load nodes through clustered linear propagation and inter-cluster weighted fusion, while incorporating correction propagation for extreme errors. The specific steps are as follows: 3.1 Clustering Calculation of Linear Combination Coefficients for Carbon Flow Rate For each generator cluster Calculate the number of generators within a cluster. For each load node The linear combination coefficient of carbon flow rate The formula is as follows:

[0052] in, For nodes The load power, For nodes The power flowing through (output of step 1). , For unit column vectors (the first one) , One component is 1, and the rest are 0). For generator The real-time power generation (the corresponding value in the cluster split power vector output in step 1); obtain the cluster For each load node linear combination coefficient matrix .

[0053] 3.2 Derivation of carbon flow rate contribution distribution parameters by clustering 3.2.1 Definition of Cluster-Level Carbon Flow Rate Contribution For each cluster Its impact on load nodes carbon flow rate contribution This is a weighted linear combination of the carbon emission intensities of all generators within the cluster, i.e.:

[0054] 3.2.2 Derivation of Parameters for Conventional Error Intervals Because all generators within the cluster Unified, satisfying the linear combination property of the Alpha stable distribution, derivation Distribution parameters within the normal error range ,as follows: (1) Stability parameters: (and cluster level) Consistent); (2) Scale parameters: ; (3) Skewness parameter: ; (4) Position parameters ( ): ; (5) Position parameters ( ):

[0055] 3.2.3 Parameter Fusion in Extreme Error Ranges For extreme error intervals, clusters Gumbel extreme value distribution parameters, in terms of cluster-level linear combination coefficients Perform weighted fusion to obtain clusters For nodes The extreme carbon flow rate error correction parameter ensures that the uncertainty of extreme errors is accurately transmitted.

[0056] 3.3 Inter-cluster weight fusion to obtain the final carbon flux mixing distribution parameters of the nodes. 3.3.1 Calculate the cluster-level contribution weight For load nodes Based on the linear combination coefficient matrix of each cluster and the nominal level of carbon emission intensity of generators ( ), calculate each cluster For nodes Carbon flow rate contribution weight The formula is as follows:

[0057] Contribution weights reflect the contribution of each cluster to the nodes The extent of the influence of carbon flow rate.

[0058] 3.3.2 Fusion of Distributed Parameters Cluster-level contribution weight The carbon flow rate contribution distribution parameters of each cluster are weighted and fused to obtain the load nodes. Final carbon flow rate mixing distribution parameters The fusion rules are as follows: (1) Stability parameters : The cluster with the largest contribution weight This decision ensures the continuity of the linear combination property. (2) Scale / skewness / position parameters: , , ; (3) Extreme error correction parameters: The extreme error correction parameters of all clusters are fused to obtain the node The extreme carbon flow rate error range eventually forms a node. The mixed distribution model of carbon flow rate (alpha stable distribution in the normal range, and Gumbel distribution correction in the extreme range) is denoted as: .

[0059] 3.4 Application Examples A two-unit, four-node system is used as an application example. The system includes one coal-fired unit and one gas-fired unit, and the carbon emission intensity of each type of unit is modeled as an Alpha stable distribution. The linear combination coefficients from the generator to the load node are calculated based on the power flow distribution matrix. The probability distribution of the load carbon flow rate at node 3 is obtained using the method described in this application, and verified through Monte Carlo simulation. The results are as follows: Figure 3 As shown. By Figure 3 It can be seen that the theoretical distribution of the carbon flow rate of the load at node 3 is basically consistent with the distribution of the Monte Carlo simulation sample, indicating that the method of this application can realize the probabilistic transfer of the uncertainty of carbon metering on the source side to the load node.

[0060] Furthermore, using the IEEE 30-bus system as an extended application example, theoretical calculations and Monte Carlo simulations were performed to compare the average carbon flow rate of multiple load nodes. The results are as follows: Figure 4 As shown. By Figure 4 It can be seen that the theoretical calculation results of each load node are consistent with the Monte Carlo simulation results, indicating that the method in this application is applicable to carbon flow rate probability tracking calculation in multi-node power grid scenarios.

[0061] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for probabilistic tracking of carbon emission distribution at power grid nodes, characterized in that, include: Step 1. Construct a clustered power flow distribution matrix, establish a linear mapping between power generation and node flow power, and use a clustering algorithm to divide the generators into several clusters based on carbon metering heterogeneity; Step 2. For each generator cluster, calibrate the basic parameters of the Alpha stable distribution in a differentiated manner, construct a cluster-level time-varying influencing factor set, train the time-varying parameter model to dynamically update the distribution parameters, and combine the Gumbel extreme value distribution to specifically model extreme carbon measurement errors, and construct a time-varying Alpha stable and Gumbel extreme value hybrid distribution model. Step 3. Based on the linear combination characteristics of the Alpha stable distribution, the distribution parameters of the contribution of each cluster to the carbon flow rate of each load node are analytically derived, and the final mixed distribution model of the carbon flow rate of the load node is obtained by passing the extreme error correction parameters and integrating them with the inter-cluster contribution weights.

2. The method according to claim 1, characterized in that, Step 1 specifically involves: constructing a clustered adaptive power flow distribution matrix and calculating the power flowing through the nodes; defining the power balance relationship for each node in the power system according to the power conservation law, and reorganizing the power balance relationship of each node into a matrix equation. Construct the power flow distribution matrix Finding its inverse yields ,pass Calculate the power vector flowing through the nodes Collect generator operating characteristic data and historical carbon metering error data, construct quantitative clustering feature vectors, and use the K-means clustering algorithm to divide all generators in the system into groups. Clusters, generator power vector Split into clusters .

3. The method according to claim 2, characterized in that, Step 2 specifically involves: clustered time-varying Alpha stable distribution modeling, and constructing a hybrid distribution model for each generator cluster. Cluster-level stability parameters are differentiated using the maximum likelihood estimation method. And calibrate the basic skewness parameters of each generator in the cluster. Basic Scale Parameters Basic position parameters For each cluster, a cluster-level time-varying influence factor set is constructed, and time-varying parameter models for skewness, scale, and location are trained, with the time-varying parameters updated in real time. , , A time-varying alpha stable distribution of carbon emission intensity for each generator was established; the cluster-level extreme carbon measurement error threshold was determined using the 3σ principle combined with the 99th percentile method. The extreme error interval is modeled using the Gumbel extreme value distribution to construct a mixed distribution model, and the effectiveness is verified by the KS test.

4. The method according to claim 3, characterized in that, Step 3 specifically involves: clustered linear transfer and inter-cluster fusion, deriving the carbon flow rate mixing distribution parameters of the load nodes: for each generator cluster... According to the inverse matrix of the power flow distribution matrix Calculate the linear combination coefficient of carbon flow rate of each generator in the cluster to each load node. The distribution parameters of each cluster's contribution to the carbon flow rate at each load node are analytically derived using the linear combination property of the Alpha stable distribution. The extreme carbon flow rate error correction parameters for each cluster are obtained by weighted fusion according to the Gumbel extreme value distribution parameters. Finally, the weights of each cluster's contribution to the carbon flow rate at each load node are calculated. The distribution parameters of each cluster and the extreme error correction parameters are integrated according to the contribution weight to obtain the final mixed distribution model of carbon flow rate of each load node.

5. The method according to claim 2, characterized in that, The power flow distribution matrix mentioned in step 1 The element is defined as: Among them, This is the power vector flowing through the nodes, with diagonal elements set to 1 and off-diagonal elements representing upstream nodes. For downstream nodes The power contribution ratio of the elements is zero; the remaining elements are zero.

6. The method according to claim 2 or 5, characterized in that, The construction of the quantitative clustering feature vector in step 1 includes: using carbon emission intensity as the target variable, and constructing a generator operation feature set based on unit thermal state characteristics, pressure and vacuum characteristics, steam and water and flue gas flow characteristics, load and energy consumption characteristics, and flue gas emission characteristics; further combining the central tendency, dispersion, skewness, and tail characteristics of historical time-series data of carbon metering errors, as well as the model specifications, range, accuracy level, sampling frequency, and operation and maintenance status of carbon metering equipment, to construct a quantitative clustering feature vector and perform normalization processing; determining the optimal number of clusters based on the silhouette coefficients corresponding to different candidate cluster numbers; updating the quantitative clustering feature vector and re-clustering when carbon metering equipment is replaced, equipment parameters are adjusted, or operation and maintenance status changes.

7. The method according to claim 3, characterized in that, The core factors of the cluster-level time-varying influence factor set mentioned in step 2 include: system load rate, generator real-time output fluctuation coefficient, carbon metering equipment operating time, and ambient temperature; for new energy unit clusters, the new energy output penetration rate is added as a time-varying influence factor; the time-varying influence factor sets of different clusters are set differently.

8. The method according to claim 3 or 7, characterized in that, The time-varying parameter models described in step 2 are determined from ARIMA, LSTM, and XGBoost models. Specifically, the ARIMA model determines its order based on the minimum AIC or BIC criterion; the LSTM and XGBoost models determine their structure and parameters based on the prediction error of the validation set; the model with the smallest overall prediction error on the validation set is selected as the time-varying parameter model for the corresponding cluster; the Bootstrap repeated sampling method is used to evaluate the time-varying parameter models, obtaining the predicted values ​​of each time-varying parameter and their corresponding confidence intervals to quantify the uncertainty of the time-varying parameter estimation.

9. The method according to claim 3 or 7, characterized in that, The personalized correction coefficient mentioned in step 2 The generator's basic parameters are determined based on the deviation between the cluster's generator operation and maintenance level and the cluster's average level; these parameters are then fine-tuned using personalized correction coefficients to obtain the final generator basic parameters: All generators within the cluster use cluster-level stability parameters. This satisfies the requirements of the linear combination theory of the stable Alpha distribution.

10. The method according to claim 3 or 7, characterized in that, The construction rule for the mixed distribution model described in step 2 is: within the normal error interval ( The time-varying Alpha stable distribution probability density function is adopted; in the extreme error interval ( The probability density function of the Gumbel extreme value distribution is used instead; a probability smoothing coefficient is introduced at the threshold junction. This achieves a continuous transition between the two distribution segments, avoiding abrupt probability changes; where PDF represents the probability density function. The probability density function is a mixed distribution. For a mixed distribution model, For probability smoothing system, This represents the probability density function of the stable Alpha distribution. Let Gumbel be the probability density function of the extreme value distribution.

11. A probability tracking system for carbon emission distribution at power grid nodes, characterized in that, include: Matrix construction module: Construct a cluster-adaptive power flow distribution matrix, establish a linear mapping between power generation and node flow power, and use a clustering algorithm to divide generators into several clusters based on carbon metering heterogeneity; The module for constructing a hybrid distribution model is as follows: Alpha stable distribution basic parameters are calibrated differently for each generator cluster, a cluster-level time-varying influencing factor set is constructed, a time-varying parameter model is trained to dynamically update the distribution parameters, and extreme carbon measurement errors are specifically modeled by combining Gumbel extreme value distribution, thus constructing a time-varying Alpha stable and Gumbel extreme value hybrid distribution model. Analytical derivation module: Based on the linear combination characteristics of the stable Alpha distribution, the distribution parameters of the contribution of each cluster to the carbon flow rate of each load node are analytically derived. By transferring the extreme error correction parameters and fusing them with the contribution weights between clusters, the final mixed distribution model of the carbon flow rate of the load nodes is obtained.

Images